Maximum F1-Score Discriminative Training Criterion for Automatic Mispronunciation Detection

Abstract

We carry out an in-depth investigation on a newly proposed Maximum F1-score Criterion (MFC) discriminative training objective function for Goodness of Pronunciation (GOP) based automatic mispronunciation detection that makes use of Gaussian Mixture Model-hidden Markov model (GMM-HMM) as acoustic models. The formulation of MFC seeks to directly optimize F1-score by converting the non-differentiable F1-score function into a continuous objective function to facilitate optimization. We present model-space training algorithm according to MFC using extended Baum-Welch form like update equations based on the weak-sense auxiliary function method. We then present MFC based feature-space discriminative training. We train a matrix projecting from posteriors of Gaussians to a normal size feature space, and add the projected features to traditional spectral features. Mispronunciation detection experiments show MFC based model-space training and feature-space training are effective in improving F1-score and other commonly used evaluation metrics. It is also shown MFC training in both the feature-space and model-space outperforms either model-space training or feature-space training alone, and is about 11.6% better than the maximum likelihood (ML) trained baseline in terms of F1-score. Further, we review and compare mispronunciation detection results with the use of MFC and some traditional training criteria that minimize word error rate in speech recognition. The experimental analysis and comparison provide useful insight into the correlations between F1-score maximization and optimization of these training criteria.

Mispronunciation Detection Leveraging Maximum Performance Criterion Training of Acoustic Models and Decision Functions

“…In this paper, we explore to learn the DNN-HMM based acoustic models, as well as the decision function, with a discriminative objective that is directly linked to the ultimate evaluation metric of mispronunciation detection. Here we take the F1-score for investigation, since it was frequently adopted as the evaluation metric in previous work on mispronunciation detection [22][23][24]. Further, in this paper, the parameters of the decision function is set to be either phone-or senonedependent when the phone-level (cf.…”

“…Yet there still are a wide array of studies that capitalize on various acoustic and prosodic cues, confidence measures and speaking-style information, to name just a few, for use in mispronunciation detection. Interested readers may also refer to [13][14][15][16][17] for comprehensive and enjoyable overviews of state-of-the-art methods that have been successfully developed and applied to various mispronunciation detection tasks.…”

Pattern Recognition in Cattle Brand using Bag of Visual Words and Support Vector Machines Multi-Class

“…Precision and Recall are measurements originated from Information Recovery and used in Classification when working with non-balanced classes. Precision is the percentage of instances that were correctly classified as positive among all of the data that were classified as positive, while Recall is the percentage of instances that were correctly classified as positive among the ones that really were positive, and F1-score is the harmonic mean between precision and recall [26]. The advantage of the F1-score is that it offers only one quality metric, facilitating a better understanding for end users.…”

MeLiF+: Optimization of Filter Ensemble Algorithm with Parallel Computing

“…The number of selected features was constant: N = 100. In order to compare our method with the old one, we used F 1 score [11] of SVM classifier.…”